the construction-development curve: evidence from a new … · the construction-development curve:...
TRANSCRIPT
Munich Personal RePEc Archive
The construction-development curve:
evidence from a new international
dataset
Girardi, Daniele and Mura, Antonio
DEPS, University of Siena, CRESME Ricerche
1 May 2014
Online at https://mpra.ub.uni-muenchen.de/64554/
MPRA Paper No. 64554, posted 26 May 2015 13:46 UTC
The construction-development curve: evidence
from a new international dataset
Daniele Girardi1 and Antonio Mura
2
1 Department of Economics and Statistics, University of Siena, Piazza San Francesco 7, I-53100
Siena, Italy 2 CRESME Ricerche, Viale Gorizia 25/C, I-00198 Rome, Italy
Corresponding author: Daniele Girardi, [email protected]. The authors are
grateful to Prof.Silvia Ferrini, Prof.Massimo Di Matteo and Prof.Tiziano Razzolini
(University of Siena) and to Robert Embleton for useful comments on earlier versions of
this paper. The usual disclaimer applies. Jel classification: L74, 010, N60 Keywords: Construction, Development, Investment, Bon‟s Curve
Published as:
Girardi, D. and Mura, A. (2014) “The Construction-Development Curve: Evidence from
a New International Dataset”, The IUP Journal of Applied Economics, Vol.XIII, No.3
Abstract
Using a new dataset of construction investment in world countries for the period 2000-
2011, this article provides novel evidence of a bell-shaped relationship between the share
of construction in GDP and economic development. The relative level of construction
activity tends to increase in developing countries, to peak during industrialization and to
decrease at a slowing pace in industrialized countries, approaching stabilization in
mature economies. The curve fits better if economic development is measured by
alternative indicators instead of per-capita GDP, namely life expectancy and an
Economic Development Index (EDI) which takes into account per capita income, life
expectancy, maternal mortality ratio and the share of agriculture in employment. On
average, the peak in construction activity is reached at a per capita income level of
almost 5,000 Euro (PPP, 2011 prices), or when life expectancy in the country has
reached around 67 years. At its peak, construction accounts for about 14% of a country‟s
GDP. The curve is robust to the inclusion of control variables. Population density,
demographic growth and credit expansion don‟t explain cross-country variation in the
share of construction in output, while there is weak evidence that a less concentrated
income distribution is positively related to the size of the construction sector.
1. Introduction
In a seminal paper in construction economics Bon (1992) proposed an inverted U-shaped
pattern for the relation between construction and economic development. The share of
construction in GDP tends to increase during the first stages of economic growth, to
stabilize in middle-income countries and to decline in advanced economies. This pattern
is often referred to as „Bon curve‟.
Providing an empirical framework to explain the level of construction activity in a
country, Bon curve could help forecast construction sector dynamics and assess whether
the size of the construction industry is in line with its long-run pattern or short-run
factors (for example a property bubble) are influencing it in a relevant way. No less
importantly, its implication that the construction sector is able to persistently „outgrow‟
the rest of the economy in developing countries is relevant for the debate about whether
or not construction should be actively supported as a driving force of growth by policy-
makers in those countries.
Empirical studies investigating Bon‟s model yielded mixed results. The most recent of
these works (Choy, 2011) argued that Bon curve doesn‟t hold across countries. In this
paper, to the contrary, we will provide evidence that Bon curve does indeed explain a
relevant fraction of cross-country variation in the share of construction in GDP. With
respect to previous studies we employ a new dataset, which allows us to measure
construction activity through gross investment instead of value added (the latter being
employed in most previous works). Furthermore, we control for three sources of
distortion that may have affected previous studies, namely non-stationarity, omitted-
variables bias and outliers.
Besides this, we refine the model proposed by Bon from two points of view. First, we
show that the curve is asymmetric with respect to its maximum. This implies that the
share of construction in GDP decreases at a slowing pace after industrialization,
approaching some kind of “plateau” in mature economies. Moreover, we take into
account a broader definition of economic development by replacing GDP per capita with
alternative indicators.
In order to be useful for forecasting and drawing more precise policy implications the
model would need to be enriched with an assessment of the other structural factors that
influence the share of construction in output. This is one of the tasks that we deal with
here. We show that population density, demographic growth and credit expansion carry
no explicative power, while there is weak evidence that a better (i.e., less concentrated)
income distribution is associated with a bigger relative size of the construction sector.
This appears to suggest that, on a microeconomic level, demand for housing exhibits a
positive and decreasing income elasticity.
The paper is organized as follows. Section 2 briefly reviews the literature on the long-run
relation between construction and development. Section 3.1 describes the data and
reports preliminary tests. In 3.2 we report and discuss results, before drawing conclusions
in section 4.
2. A critical review of literature
There are three main strands in the literature on the economic role of construction. The
first one studies the relationship between construction and development. The second tries
to assess whether construction investment leads GDP growth (De Long and Summers,
1991; Ball and Wood, 1996; Chang and Nieh, 2004). The third one employs input-output
tables to study the role of construction in a national economy (Bon and Pietroforte, 1990;
Pietroforte and Gregori, 2003). This paper is concerned with the first strand.
Early seminal papers investigating the role of construction in economic development are
the ones of Strassmann (1970), Turin (1974), Drewer (1980), Wells (1985) and Bon
(1992). They tried to assess whether “the construction sector, like agriculture or
manufacturing, follows a pattern of change that reflects a country‟s level of
development”, as Strassmann (1970) put it. The most influential was probably the work
of Bon (1992). He argued that the share of construction in GDP follows a bell-shaped
pattern: it tends to increase during the first stages of economic growth, to stabilize in
middle-income countries and to decline in advanced economies. Following some recent
literature (Ruddock and Lopes, 2006; Choy, 2011) we can refer to this pattern as Bon
curve1 or, alternatively, the Construction-Development Curve (hereafter CDC).
The intuition behind Bon curve is that earlier stages of growth are characterized by
intense processes of urbanization, demographic growth, creation of basic infrastructures
and construction of industrial plants. Thus, the construction sector tends to grow faster
than the rest of the economy during this phase, increasing its share in output. In later
stages, as these processes reach maturity and start slowing down, growth in construction
investment tends to slow down with respect to the overall economy. This theory is
consistent with the empirical finding that both fixed capital formation and the share of
durable physical assets in investment tend to be larger in developing countries and to
1 It is fair to note, however, that well before the work of Bon this bell-shaped relation had already been
highlighted by Strassmann (1970).
decline in advanced economies (Bon, 1992; Maddison, 1987) and with the S-shaped
relationship that has been found between urbanization and economic growth (Berry,
1973).
Bon curve appears compatible with a Lewisian view of economic development.
According to the Lewis model (Lewis, 1954) the ultimate cause of economic
development is the shift of workforce from a low-productivity subsistence sector to a
capitalist advanced industrial sector. Owing to underemployment, expanding capitalist
industries enjoy an unlimited supply of unskilled labour at a subsistence wage. As long
as profits are reinvested, the process is self-sustaining: increasing investment results in a
further expansion of the industrial sector, which allows further extraction of
underemployed workforce from agriculture. Within this theoretical framework there are
reasons to expect the construction sector to grow faster than the rest of the economy in
developing countries. The flow of labour from subsistence agriculture to industry
determines urbanization. Furthermore the construction industry is particularly good in
extracting underemployed workforce, because it typically requires a larger amount of
unskilled labour and a less sophisticated level of entrepreneurship than other industrial
sectors.
Some works have used cross-section analysis to study the relation between construction
and economic development, with mixed results. In the most recent of these studies, Choy
(2011) grouped 205 countries into four groups on the basis of per capita income and
compared the averages of the construction‟s share of GDP (measured by gross value
added) in the four groups in the period 1970-2009. For most years he found the average
share of construction in output to be slightly higher in rich countries than in middle-
income economies. Moreover he performed time-series estimations of Bon curve for each
country over the period 1970-2009, finding the curve to be significant in most rich
countries. On the basis of these results he argues that Bon model holds within most
developed countries over time but not across countries at a given time.
Overall, empirical works on the relation between construction and development have
been rather descriptive. Most of them (with the partial exception of Crosthwaite, 2000)
group countries into four categories depending on per capita income and then calculate a
simple average of the construction‟s share of GDP in each group. In some cases, these
average values are taken over multi-year periods, so they are likely to be biased because
of non-stationarity of the data and changes in the composition of groups. Omitted
variable-bias and outliers are two further potential sources of distortion: none of these
studies checked for the presence of possible outliers in their samples; none of them tried
to assess whether other factors influence the share of construction in GDP beyond the
level of per-capita income, so the robustness of Bon curve to the inclusion of control
variables has never been tested.
3. Empirical evidence for the period 2000-2011
In what follows we test empirically Bon curve. The first part of this section describes the
dataset and reports the preliminary analysis of the data. In the second part results are
presented and discussed.
INSERT TABLE 1 ABOUT HERE
3.1 Overview of the data and preliminary tests
Statistical information about construction investment in world countries comes from the
Simco database2, managed by Cresme Ricerche
3. Simco gathers data from official
national sources, covering 148 countries (accounting for 99% of world GDP, 98% of
world population and 98% of world surface) for the 2000-2011 period.
Global construction activity reached 5,600 billion Euro in 2011, displaying in real terms
a 48% increase over 2000. Basic infrastructures represent the main component of
construction investment in African and South American countries, while Europe is
characterized by a major incidence of residential activities, especially those related to
renewal and maintenance. Non-residential investment is the main source of construction
investment in India and Russia. It is also dominant in the United States, where residential
activity has not yet recovered after the burst of a huge housing bubble in the late 2000s.
The composition of global construction investment by sector and macro-area in 2011 is
summarized in Table 1.
As shown in Fig. 1, construction investment as a share of GDP has been increasing in
Africa, Asia and South America. However, a peak in the share of construction in world
output has been reached in 2006, when construction peaked in Europe (12.2%) and North
America (9.8%).
INSERT FIGURE 1 ABOUT HERE
2 Cresme has kindly allowed the authors to provide the dataset used in this work by email to anyone who
requests it for academic purposes
3 An Italian non-profit institution whose aim is to produce research on the building sector, the real estate
market and their impact on land transformation
The boom and bust in the housing market of some industrialized countries has of course
affected the data. Other major episodes which happened during our sample period and are
likely to have affected the data for some economies were two historical peaks in
commodity prices (and a steep fall in between) and some major episodes of civil turmoil
in certain countries. In some economies, especially the smaller and less diversified ones,
the short-run fluctuations caused by these idiosyncratic shocks in our variable of interest
(i.e. construction investment as a share of GDP) may have been huge enough to obscure
and overwhelm the long-run tendency that we are trying to investigate. For this reason
we used standard procedures4 in order to detect outliers.
Three groups of outliers emerged. First, countries that have experienced and anomalous
level of construction activity due to a strong housing bubble: Spain, Ireland and Iceland.
Certainly other western economies have also been affected by housing speculation in the
same period, most importantly the US and the UK, but in those countries the share of
construction investment in GDP was relatively unaffected, because of their larger and
more diversified economies (Fig. 1). Second, small oil exporting countries, such as Qatar,
Bahrain and Arab Emirates. In these countries both construction investment – a large part
of which is accounted for by oil-related infrastructures – and GDP are heavily affected by
oil revenues and commodity market dynamics. A different case is the one of Singapore,
which has been excluded because of its particular city-state nature. Third, countries
which data display evidence of inconsistency, due to scarcity and/or inconsistency of
statistical information. In most cases we acknowledged that different sources provided
very different estimates (this is the case of some African countries, Albania, North Korea
4 More precisely, we applied a simple k-means clustering method in a bi-dimensional space defined by
construction investment as a share of GDP and per-capita GDP.
and Vietnam). In the case of Afghanistan, the lack of reliable statistical information is
due to the armed conflict that the country has experienced during our period of interest.
Overall, this analysis led us to exclude 20 observation units5, reducing our sample from
148 to 128 countries. After removing those observations our sample still accounts for
94% of world population and 95% of world GDP.
Model specification
Visual inspection of the data suggests that the relation between construction investment
and economic development is indeed inverted U-shaped but asymmetric with respect to
the maximum. However a symmetric bell-shaped pattern seems to be recovered after
logarithmic transformation of both variables (Fig. 2). A preliminary cross-section
analysis confirms this intuition. We estimate a quadratic relation between the share of
construction in output and GDP per capita for each year in the sample using three
different specifications: taking both variables in absolute values („lin-lin‟ model); taking
logarithms of the independent variable („lin-log‟); taking logarithms of both variables
(„log-log‟). The log-log model appears to be the best choice in terms of parameters
significance (Table 2).
INSERT FIGURE 2 ABOUT HERE
INSERT TABLE 2 ABOUT HERE
5 Spain. Ireland, Iceland, Bahrain, Qatar, Libya, UAE, Albania, Vietnam, Singapore, Angola, Somalia,
Eritrea, Lesotho, Guinea Bissau, Zambia, Democratic Republic of Congo, Afghanistan, North Korea,
Tajikistan.
Estimation strategy
Different estimators can be employed in a panel setting, depending on the characteristics
of the data. The nature of our panel is such that a between-groups estimator (i.e. a cross-
section regression using the average value of the data over the sample period for each
unit of observation) appears to be a natural choice. In our sample variability between
countries is in practice the only source of relevant information, since we have countries
with very different levels of economic development, while the time dimension of the
panel is far too short to allow us to observe different stages of development within each
single country. In such a short time span the level of economic development of a country
can be considered almost as a time-invariant factor, so there would be no point in trying
to exploit within-groups variability to estimate its effect.
Stationarity test
To test stationarity, we perform separate estimates of the CDC (eq.1) for each year in the
sample period 2000-2011. Estimated yearly coefficients (plotted in Fig. 3) are suggestive
of a break in 2006. After 2006 we observe what seems to be a structurally higher level of
β1 and a lower level of β2, implying that the peak in construction activity level is reached
on average at a lower level of income. The shift resulted from both an increase in
construction activity level in developing countries and a decrease in mature economies.
While in advanced countries construction has been hit harder than other economic
sectors during the crisis, in some major developing countries (most notably the so-called
BRICs) huge public infrastructure-related investments were put in place as an attempt to
boost growth after the downturn. In order to take into account this structural change and
avoid a non-stationarity bias in the estimated coefficients we split our sample into two
sub-periods, 2000-2006 and 2007-2011.
INSERT FIGURE 3 ABOUT HERE
3.2 Results
On the basis of the preliminary analysis reported above, we estimate the following
equation in our sample:
2
21 )log()log()log( XXY (eq.1)
Y is construction‟s share in output, X is a proxy for development, ε is a random shock
and the bar over a variable means that its average over the sample period is taken. Bon‟s
hypothesis would imply β1 > 0 and β2 < 0.
At first we employ purchasing-power parity (PPP) per capita GDP as a proxy for
economic development. Results are reported in Table 3. In 3a the share of construction in
domestic output is measured by construction investment as a share of GDP, while in 3b it
is measured through the construction‟s share of value added. In both cases the inverted
U-shaped relationship holds: coefficients have the expected sign and are significant at
any conventional level in both periods.
When employing investment the R2 of the model increases significantly in the second
sub-period (from 8.7% to 13.6%), while when employing value added the goodness of fit
of the model does not change between the two sub-periods (R2 is 9.6% in both).
Furthermore in the specification using value added the difference in the estimated
coefficients between the two sub-periods is lower. These differences are probably due to
the fact that construction investment was more volatile than construction value added in
our period of interest. For our purposes, however, investment appears to be a more
suitable indicator, since it represents a broader measure of the weight of construction in a
national economy, taking into account not only the net product of the sector but also its
demand for intermediate goods.
From these results we can estimate the average level of income at which the share of
construction in GDP tends to reach its maximum level. Investment in construction as a
share of GDP reaches a peak of 12% at an income level of 6,500 Euro per capita in the
2000-2006 period, while in 2007-2011 it peaks at 14%, with per capita GDP at 4,900
Euro (more or less China‟s level of income). The construction‟s share of value added
reaches a maximum of 5.6% at a level of per capita income of 7,900 Euro in 2000-2006,
while in 2007-2011 the peak is at 6.4% at a level of per capita income of 6,500 Euro (at
PPP and in 2011 prices).
INSERT TABLE 3 ABOUT HERE
3.2.1 Replacing GDP with alternative measures of development
While at the time of Strassmann‟s 1970 pioneering paper growth and development were
almost universally considered as synonyms, nowadays there is an extensive literature
showing that GDP is a rather poor and mono-dimensional measure of economic
development and stressing the need for broader and more comprehensive indicators
(Stiglitz et al., 2010). We take these considerations into account by re-estimating eq.1,
replacing GDP per capita with alternative indicators of economic development.
INSERT TABLE 4 ABOUT HERE
At first, we try using the Human Development Index (HDI) calculated by the UNDP,
which takes into account per-capita income, average and expected years of schooling and
life expectancy at birth6. As shown in Table 4, the use of the HDI instead of per capita
GDP does not improve the fitness of our model: the relation is not significant in the first
sub-period (2000-2006), while in the second (2007-2011) coefficients are statistically
significant but R2 is lower than the one obtained by using per capita income (9.8% versus
13.6%).
We then re-estimate the model employing a broader Economic Development Index
(EDI)7 which takes into account per capita income, life expectancy at birth, the share of
workforce employed in agriculture and the maternal mortality ratio. The differences
between our EDI and the HDI reflect the fact that we did not aim at building a measure of
well-being but, more simply, an indicator of economic development. The use of the EDI
as a proxy for development yields a significantly higher explanatory power in both sub-
periods. R2 is 14.6% in the first subsample (2000-2006) and 20.2% in the second (2007-
2011), against 8.7% and 13.6% that we obtained by using per capita income (Table 4).
6 See UNDP (2011, pp. 168-169) for details. The HDI was downloaded from the UNDP at
http://hdr.undp.org/en/statistics/hdi
7 We built the EDI by principal-component analysis. It is a weighted geometric mean of the principal
components, with weights proportional to the explained variance. Variables were taken from the World Bank
Database.
However, and rather interestingly, an even better approximation to the actual data is
obtained if we use life expectancy at birth alone as the proxy for economic development.
R2 increases to 16.9% in the first subsample (2000-2006) and to 24.2% in the second
(2007-2011). This could be due to the fact that life expectancy is closely related to the
level of economic development, without being distorted by country-specific factors that
could instead influence other indicators. In other words, probably no country has an
idiosyncratic characteristic, unrelated to economic development, which makes life
significantly longer for its inhabitants. In the first sub-sample (2000-2006) construction
investment as a share of GDP tends to reach a maximum of 12% when life expectancy is
68.5 years, while in the second sub-sample (2007-2011) it reaches a maximum of 14.4%
in correspondence with a life expectancy of 66.8 years (Table 4).
3.2.2 Inclusion of control variables
We then enrich the model by including some further independent variables8. There are
two reasons for doing so. On the one hand, we want to assess whether other measurable
factors influence the weight of construction in GDP, beyond the level of economic
development. On the other hand, the inclusion of control variables allows us to test the
robustness of the CDC.
Physical characteristics, such as surface, population and population density, are natural
candidates. Land is a fundamental production factor for the construction industry, while
8 Here we use our EDI as the proxy for development but results don‟t change if we use per capita GDP or life
expectancy. All variables used as controls were downloaded from the World Bank Database
the number of inhabitants is a measure of potential demand for buildings and
infrastructures. However, since our dependent variable is a relative measure of
construction activity (i.e., construction investment as a share of GDP), which does not
depend on the economy‟s size, the most appropriate variable to include in our model is
probably population density9.
Since a large part of construction activity consists in what Simon Kuznets famously
called “population-sensitive capital formation” (Kuznets, 1961), another natural
candidate for explaining cross-country variation in the share of construction in GDP is
the growth rate of population.
A further factor that may help explain construction activity level is income distribution.
A more equal distribution of income could result in a larger share of population which
can afford decent housing and in a greater availability of public services, thus fostering
demand for buildings and public infrastructures. An analogous way of seeing this relation
is that of considering housing as a normal good with decreasing income elasticity: low-
and medium- income families are likely to spend a larger share of their income on
housing (through purchases of first homes, rents and mortgages) than high income
families. We use the Gini Index as a proxy for income distribution.
9 It is not clear, however, which sign should be expected from this relation. On the one hand, higher
population density could be associated, ceteris paribus, with greater demand for residential buildings and
infrastructures; on the other hand, lower population density could result in a greater availability of land for
construction projects.
Credit growth may also be thought to affect the share of construction in GDP, through
mortgage loans and loans to construction firms. We thus include domestic credit to the
private sector (as a % of GDP) in the regression.
As shown in Table 5, the coefficients of surface, population, population density,
population growth and credit/GDP are not statistically significant. It is somehow
surprising to note that countries with higher demographic growth don‟t exhibit a higher
share of construction investment in output. The coefficient of the Gini Index is significant
and has the expected (negative10
) sign in the second sub-period, but it is not significant in
the first. A better (i.e. less concentrated) distribution of income appears to be associated
with a higher level of construction activity. According to our estimates in the period
2007-2011 an increase in the Gini Index by one standard deviation is associated, on
average, with a 0.09 decrease in the natural logarithm of the share of construction in
output. This implies that an increase in the Gini Index from the third to the second
quintile (which in our case means a 10% increase in the index) is associated with a 4.4%
decrease in the expected value of the share of construction in output.
The CDC is robust to the inclusion of the abovementioned control variables. Both the
linear and the quadratic coefficient of the Economic Development Index remain
statistically significant at any conventional level, and with the expected sign. According
to our estimates, as the EDI passes from its lower value in the sample (which is 57.7 and
corresponds to Burundi) to the value that maximizes the curve (that is 81, near to the
value assigned to Philippines, Georgia or Moldova) the expected value of the share of
10 In interpreting the sign of the coefficient, one has to remember that the higher the Gini Index, the more
concentrated the distribution of income.
construction in GDP triples, while as we pass from the peak of the curve to the maximum
value of the EDI (which is 99.4 and corresponds to Norway) the expected value of the
share of construction in GDP decreases by 34.7%. This means that, on average, in the
increasing part of the curve (i.e., in developing countries) a 10% increases in the EDI is
associated with a 38% increase in the share of construction in GDP, while in the
decreasing part of the Curve (i.e., in mature economies) a 10% increase in the EDI is
associated with an 18% decrease in the relative level of construction activity.
INSERT TABLE 5 ABOUT HERE
INSERT FIGURES 4,5 AND 6 ABOUT HERE
3.2.3 Robustness checks
In order to check the robustness of our results to the method employed to detect outliers,
we re-estimated the model on two different subsamples.
First, we completely ignored outliers, leaving all available observations in the sample.
Estimated coefficients maintain the same sign in the model with per-capita GDP as a
proxy for development, but with lower statistical significance. They remain highly
significant, however, in the second sub-period in the variant with construction‟s share of
value added as the dependent variable and in the one with life expectancy as a proxy for
development, and in both sub-periods in the model employing the EDI (although with a
lower R2).
Second, we excluded only countries which data appears to display some inconsistency
(very different estimates from different sources or suspiciously large changes between
years), keeping in the sample countries which experienced huge housing bubbles and
small oil exporters. So we excluded only Afghanistan, Albania, North Korea, Vietnam,
Somalia, Eritrea, Lesotho, Guinea Bissau, Zambia and Democratic Republic of Congo.
All results remain practically identical to the ones presented in previous sections and R2
is only slightly lower.
We also tried entering control variables in logarithms, obtaining no significant change in
results. On the basis of these trials – whose coefficients and test-statistics are omitted
here for brevity but can be requested to the authors – our findings appear to be rather
robust, since only including very low quality data partly perturbs results.
4. Conclusions
We have used panel data for world countries for the period 2000-2011 to provide
evidence of a bell-shaped relationship between construction activity and economic
development, consistent with the theory proposed by Bon (1992). The relation gets
stronger after logarithmic transformation of the data (Table 2). This implies that the
curve is asymmetric with respect to its maximum: the size of the construction sector
tends to increase in developing countries, to peak in newly industrialized economies and
to decline at a slowing pace afterwards, approaching stabilization in the most advanced
economies.
We have also found that the curve fits better when employing alternative indicators to
measure the level of economic development instead of per capita GDP. This supports the
intuition that the size of the construction sector is not just a function of per capita output,
but is related to broader socio-economic trends which are intimately linked with
economic development, namely urbanization, industrialization and creation of basic
infrastructures. In particular, we have found that the model fits better when economic
development is measured through an index (EDI) composed of per capita income, life
expectancy, maternal mortality ratio and the share of agriculture in employment.
However, and rather interestingly, we have obtained an even better fit to the data when
using life expectancy alone as the proxy for development. A possible explanation is that
life expectancy at birth is not distorted by country-specific factors, unrelated to
development, which could instead influence the empirical distribution of other indicators.
The curve is robust to the inclusion of control variables. Population density, demographic
growth and credit availability don‟t explain cross-country variation in the share of
construction in output. However, we find weak evidence that a better (i.e., less
concentrated) income distribution is positively related to the size of the construction
sector, suggesting that low and medium income families tend to spend a larger share of
their income on housing. Apart from this, Bon curve appears the only factor explaining
cross-country variation in the share of construction in output, at least among the variables
that we were able to include in the model.
Our findings carry two main policy implications. First, in the medium-run the
construction sector is able to grow faster than the rest of the economy in developing
countries, acting as a driving force of growth. Since the share of construction in output is
one of the variables that policy-makers usually monitor in order to spot possible
overheating in real estate markets, the second policy implication is that different criteria
must be employed in judging the value of this indicator in developing and mature
economies, given that an higher and increasing weight of construction is physiological in
industrializing countries.
A question that arises almost naturally is whether the bell-shaped pattern holds within all
sub sectors of construction or it is the result of different dynamics experienced by
housing, infrastructures and non-residential buildings or by new buildings as opposed to
renewal and maintenance activities. Intuition and a descriptive overview of the dynamics
observed in major countries – see for example Euroconstruct (2012) – would for example
suggest that the bell-shaped relationship could be determined by new buildings, while the
incidence of renewal and maintenance activities may be linearly related to economic
development (and so could explain the tendency toward stabilization in mature
economies). These questions may inspire further empirical work. The most challenging
task is probably going to be related to the current scarce availability of data on the
composition of construction investment in most developing countries.
References
Ball, M. and Wood, A. (1996) “Does building investment affect economic growth?”,
Journal of Property Research, Vol.13, 99-114.
Berry, B.J.L. (1973) “The Human Consequences of Urbanization: Divergent Paths in the
Urban Experience of the Twentieth Century”, Macmillan, London, 1973.
Bon, R. (1992) “The future of international construction”, Habitat. Int., Vol.16, 119–128.
Bon, R. and Pietroforte, R. (1990) “Historical comparison of construction sectors in the
United States, Japan, Italy and Finland using input-output tables ”, Construction
Management and Economics, Vol.8, 233–247.
Chang, T. and Nieh, C. (2004) “A note on testing the causal link between construction
investment and economic growth in Taiwan”, Journal of Asian Economics, Vol.15, 591–
598.
Choy, C. F. (2011) “Revisiting the „Bon curve”‟, Construction Management and
Economics, Vol.29, 695–712.
Crosthwaite, D. (2000) “The global construction market: a cross-sectional analysis”,
Construction Management and Economics, Vol.18, 619–627.
De Long J.B. and Summers L.H. (1991) “Equipment investment and economic growth”,
The Quarterly Journal of Economics, Vol.106, 445–502.
Drewer, S. (1980) “Construction and development: A new perspective”, Habitat. Int.,
Vol.5, 395–428.
Euroconstruct (2012) “Country Report; 74th Euroconstruct Conference, Munich,
December 2012”, Ifo Institute, Munich
Kuznets, S (1961) “Capital in the American Economy: Its Formation and Financing”,
Princeton University Press, available on-line at http://papers.nber.org/books/kuzn61-1
Lewis, W.A. (1954) “Economic Development with Unlimited Supplies of Labour”,
Manchester School of Economic and Social Studies, Vol.22, 139–191.
Maddison, A. (1987) “Growth and Slowdown in Advanced Capitalist Economies:
Techniques of Quantitative Assessment”, Journal of Economic Literature, Vol.25, 649–
698.
Pietroforte, R. and Gregori, T. (2003) “An input-output analysis of the construction
sector in highly developed economies”, Construction Management and Economics,
Vol.21, 319–327.
Ruddock L. and Lopes J. (2006) “The construction sector and economic development:
the „Bon curve”‟, Construction Management and Economics, Vol.24, 717–723.
Stiglitz J.E., Sen A. and Fitoussi J. (2010) “Report by the Commission on the
Measurement of Economic Performance and Social Progress”, available at
http://www.stiglitz-sen-fitoussi.fr/documents/rapport_anglais.pdf
Turin D. (1974) “The Role of Construction in Development Strategies”, International
Development Review, Vol.16
UNDP (2011) “Human development Report 2011”, Palgrave MacMillan, New York
Wells J. (1985) “The role of construction in economic growth and development”,
Habitat. Int., Vol.9., 55–70.
Appendix
List of countries in the sample and Economic Development Index (EDI) ranking
TABLES
Table 1: Investment in construction by sector and macro-area in 2011 (Billion Euros)
Residential Non-residential Infrastructures Total
Investment Share Investment Share Investment Share Investment Share
Asia 737 29.6% 819 32.8% 937 37.6% 2, 493 100%
Europe 667 42.4% 522 33.2% 386 24.5% 1, 575 100%
N.America 304 33.1% 329 35.9% 285 31.1% 918 100%
S.America 117 35.5% 87 26.3% 125 38.1% 329 100%
Africa 38 26.1% 35 24.0% 73 49.9% 145 100%
Oceania 43 26.9% 30 19.0% 86 54.1% 159 100%
World 1,906 33.9% 1,822 32.4% 1,897 33.7% 5,619 100%
Source: Cresme/Simco (2012)
Table 2: Average results of yearly cross-section analyses (2000-2011)
lin-lin lin-log log-log
Parameters P-Value Parameters P-Value Parameters P-Value
Constant 0.117 0.000 -0.460 0.124 -8.240 0.000
GDP 0.000 0.235 0.142 0.046 1.431 0.011
GDP2 0.000 0.122 -0.008 0.052 -0.083 0.016
R2 3.7% 6.9% 9.9%
Notes: Dependent variable: share of construction in output; GDP = per-capita GDP
Table 3: Estimation of the CDC with GDP per capita as a proxy for development
a)Dep. variable: log(Investment in construction as a share of GDP)
2000-2006 2007-2011
Coefficient P-value Coefficient P-value
Constant -2.61 0.12 -5.44***
0.003
log(GDP per capita) 1.16***
0.006 1.90***
2.3*10-5
log(GDP per capita)2 -0.07
*** 0.009 -0.11
*** 2.8*10
-5
Y max 12.0% 14.1%
X max 6,509 4, 906
Regression statistics N = 128, F-Stat = 6.0, R2 = 8.7% N = 128, F-Stat = 9.8, R2 = 13.6%
b)Dep. variable: log(Construction‟s share of value added) 2000-2006 2007-2011
Coefficient P-value Coefficient P-value
Constant -3.10* 0.07 -4.51
** 0.02
log(GDP per capita) 1.07***
0.01 1.45***
0.002
log(GDP per capita)2 -0.06
** 0.02 -0.08
*** 0.004
Y max 5.6% 6.4%
X max 7, 931 6, 501
Regression statistics N = 127, F-Stat = 6.6, R2 = 9.6% N = 127, F-Stat = 6.6, R2 = 9.6%
Table 4: Estimation of the CDC with alternative measures of economic development Dep.variable: log(Investment in construction as a share of GDP)
Human Development Index (HDI)
2000-2006 2007-2011
Coefficient P-value Coefficient P-value
Constant 2.42***
8.2*10-37
2.21***
3.0*10-39
log(HDI) -0.16 0.73 -1.47***
0.005
log(HDI)2 -0.41 0.25 -1.44
*** 0.001
Ymax 11.4% 13.4%
Xmax 0.82 0.60
Regression statistics N = 127, F-Stat = 5.7, R2 = 8.3% N = 127, F-Stat = 6.7, R2 = 9.8%
Economic Development Index (EDI)
2000-2006 2007-2011
Coefficient P-value Coefficient P-value
Constant -110.5***
0.003 -179.7***
2.0*10 -6
log(EDI) 50.9***
0.002 83.0***
1.7*10 -6
log(EDI)2 -5.73
*** 0.003 -9.43
*** 2.0*10
-6
Ymax 12.1% 14.5%
Xmax 84.7 81.2
Regression statistics N = 123, F-Stat = 10, R2 = 14.6% N = 123, F-Stat = 15, R2 = 20.2%
Life Expectancy at birth
2000-2006 2007-2011
Coefficient P-value Coefficient P-value
Constant -66.6***
0.004 -144.4***
9.1*10 -7
log(Life Expectancy) 32.7***
0.004 70.0***
8.0*10 -7
log(Life Expectancy)2 -3.87
*** 0.005 -8.32
*** 1.0*10
-6
Ymax 12.0% 14.4%
Xmax 68.5 66.8
Regression statistics N=127, F-Stat = 12.6, R2 = 16.9% N = 127, F-Stat = 19.8, R2 = 24.2%
Table 5: Estimation of the CDC with control variables
Dep.variable: log(Investment in construction as a share of GDP)
2000-2006 2007-2011
Coefficient P-value Coefficient P-value
Constant -116.0**
0.01 -198.9***
8.9*10-6
log(EDI) 53.3**
0.01 92.0***
8.5*10 -6
log(EDI)2 -6.0
** 0.01 -10.5
*** 1.0*10
-5
Gini Index -0.04 0.43 -0.09**
0.04
Population 0.03 0.53 0.01 0.74
Surface -0.05 0.25 -0.02 0.66
Population Density 0.07 0.64 -0.13 0.37
Population Growth 0.35 0.67 0.002 0.98
Credit/GDP 0.01 0.87 0.03 0.57
Regression statistics N=119, F=3, R2=18.2%, Adj.R2=12.2% N= 119, F=4, R2 = 24.0%,Adj.R2=18.4%
Note: All variables except the EDI were standardized
FIGURES
Figure 1: Investment in construction as a share of GDP
Source: CRESME/Simco (2012)
Figure 2: Country distribution with respect to GDP per capita and construction’s share in output. Linear (top panel) and logarithmic (bottom panel) scale.
Source: Cresme/Simco (2012)
Figure 3: Structural break (estimated yearly coefficients and confidence intervals)
Figure 4: Construction-Development Curve with per capita GDP as a proxy for development (2007-2011)
Figure 5: Construction-Development Curve with the EDI as a proxy for development
(2007-2011)
Figure 6: Construction-Development Curve with life expectancy as a proxy for
development (2007-2011)